1先猜，再看答案Guess first, then reveal

地球上每 100 個人，有幾個在用生成式 AI？Out of every 100 people on Earth, how many use generative AI?

不要查資料，憑你的感覺拉滑桿。你身邊的人，可能不是全世界的樣子。No searching. Go with your gut. The people around you may not look like the rest of the world.

50%
10%

使用率Usage

付費率（0–30% 尺度）Paying (0–30% scale)

1 點 ≈ 330 萬人，2,500 點 = 83 億人1 dot ≈ 3.3M people, 2,500 dots = 8.3B 每人只算在最進階的一類Each person counted once, in their most advanced group
2

把各家 chatbot 用戶數加起來，會發生什麼事？What happens if you add up every chatbot's users?

點選下面的 chatbot，看加總跟「去重後的真實人數」差多少。注意每個數字底下的指標和日期。Select chatbots below and compare the total with the deduplicated number of real people. Watch the metric and date under each figure.

WAU = 每週活躍用戶，MAU = 每月活躍用戶。來源：各家官方公布、QuestMobile、SpaceX S-1，經 Gradually AI 整理。WAU = weekly active users, MAU = monthly active users. Sources: official company figures, QuestMobile, SpaceX S-1, compiled by Gradually AI.

你選的加總Your total
0
去重後 2.42B 人Deduplicated 2.42B
上網人口 ≈ 6.1B（推算）Online ≈ 6.1B (derived)
世界 8.3BWorld 8.3B
3

同一個國家，為什麼有好幾個「使用率」？Why does one country have several “usage rates”?

選一個國家，比較不同單位量出來的數字。黃色列是定義不一樣的地方。Pick a country and compare numbers from different organizations. Yellow rows are where the definitions differ.

先自己想：差距可能從哪裡來？想好再打開Think first: where could the gap come from? Open when ready
  • 算誰：年齡範圍不同（16+、16–74、15–64），分母就不同。Who counts: different age ranges (16+, 16–74, 15–64) mean different denominators.
  • 算什麼：「任何 AI」包含導航、推薦系統；「生成式 AI」「chatbot」範圍比較窄。What counts: “any AI” can include maps and recommendations; “generative AI” and “chatbots” are narrower.
  • 時間窗：「曾經用過」一定比「過去 3 個月用過」高。Time window: “ever used” will always be higher than “used in the last 3 months.”
  • 怎麼量：問卷靠受訪者自己回答；Microsoft 用匿名化的系統資料，再依裝置市占、上網率和人口結構調整推估。兩種方法各有盲點。How it's measured: surveys rely on self-report; Microsoft uses anonymized system data adjusted for device share, internet access and demographics. Each has blind spots.
  • 什麼時候量：AI 使用率一年內可以翻倍，差半年就可能差很多。When: usage can double within a year, so six months apart matters.
4

你是編輯：這些標題可以直接刊嗎？You're the editor: can these headlines run?

用前面看到的資料判斷。每則選一個，選完會顯示理由。Use what you've seen above. Pick one verdict per headline to see the reasoning.

5

小組討論Group discussion

  1. 回到第 1 題：你猜的數字比實際高還是低？如果你身邊 100 個人畫成點陣圖，顏色分布會像全球嗎？為什麼？Back to step 1: did you guess high or low? If you drew the 100 people around you as dots, would the colors look like the world's? Why?
  2. 如果要設計一份「東華學生 AI 使用率」問卷，你會怎麼定義「使用」？時間窗要設多長？只問一題夠嗎？If you designed a survey on AI use among our students, how would you define “use”? What time window? Is one question enough?
  3. 全球有 2.2B 人完全離線。生成式 AI 會縮小還是擴大數位落差？花蓮的偏鄉和都市之間呢？2.2B people are fully offline. Will generative AI narrow or widen the digital divide? What about rural and urban Hualien?
  4. 請 AI 幫你查「台灣 AI 使用率」，它給的數字是哪個單位、哪一年、哪種定義？它有主動告訴你嗎？Ask an AI for “Taiwan's AI usage rate.” Which organization, which year, which definition does its number use? Did it tell you without being asked?

資料來源Sources

整理：2026 年 9 月。數字會持續變動，上課前建議回原始來源確認。Compiled September 2026. Figures change quickly; check the original sources before class.